ToW: Thoughts of Words Improve Reasoning in Large Language Models
This addresses the issue of reasoning and hallucination in large language models, offering a task-agnostic improvement, though it is incremental as it builds on existing pre-training methods.
The authors tackled the problem of factual hallucination and inefficient learning of implicit reasoning in next-word prediction by introducing a training-time data-augmentation method called thoughts of words (ToW), which improved reasoning performance by 7% to 9% on average and reduced hallucination by up to 10%.
We introduce thoughts of words (ToW), a novel training-time data-augmentation method for next-word prediction. ToW views next-word prediction as a core reasoning task and injects fine-grained thoughts explaining what the next word should be and how it is related to the previous contexts in pre-training texts. Our formulation addresses two fundamental drawbacks of existing next-word prediction learning schemes: they induce factual hallucination and are inefficient for models to learn the implicit reasoning processes in raw texts. While there are many ways to acquire such thoughts of words, we explore the first step of acquiring ToW annotations through distilling from larger models. After continual pre-training with only 70K ToW annotations, we effectively improve models' reasoning performances by 7% to 9% on average and reduce model hallucination by up to 10%. At the same time, ToW is entirely agnostic to tasks and applications, introducing no additional biases on labels or semantics.